Fueling the Future: Optimizing Nuclear Core and Fuel Designs

The design of the reactor core and its constituent fuel assemblies plays a critical role in determining the overall performance of a nuclear reactor. In order to generate electricity as safely and efficiently as possible in terms of cost, resources, and environmental impact, these designs must be thoroughly optimized, all while considering human and computational resources. Design is a complex, iterative process that involves exploring and refining configurations through high-fidelity simulations, with the goal of improving safety, efficiency, and performance. However, such simulations are computationally expensive, making exploration of the vast design space time-consuming and resource-intensive. This motivates the development of faster approaches to support the design process. This thesis, conducted at Vattenfall Nuclear Fuel, explores the application of Artificial Intelligence (AI) for in-core fuel management through two main components. First, it investigates where and how AI can be applied in in-core fuel management design workflows by reviewing current research and consulting domain experts. Secondly, it presents Machine Learning (ML) implementations of neural network models aimed at accelerating part of the design evaluation process. Using a fuel assembly design with the arrangement and percentages of uranium enrichment levels as input, the ML models are trained to predict key reactor parameters, including the infinite multiplication factor (k∞) and power distribution. Additionally, the internal peaking factor (Fint) can be derived from the predicted power distribution. The results demonstrate that the models can predict these key parameters with promising accuracy while being significantly faster than traditional simulation methods. This suggests that ML has the potential to augment or partially replace simulation in specific stages of the nuclear design workflow. The work contributes to bridging the gap between advanced ML methods and traditional nuclear engineering practices, laying a foundation for further integration of ML into in-core fuel optimization.